face_match_kit 0.1.0 
face_match_kit: ^0.1.0 copied to clipboard
Published 16 days ago • 
nexdark.com
nexdark.comOn-device face detection, guided enrollment, liveness challenges, and 1:1 verification for Flutter.
Unreleased #
- Added
CaptureFlowPolicy.simpleVerification: single straight look for verification, no turns. - Aligned live pose windows with the still-enrollment gate (front ±12°, sides 10–42°) and widened the frame-gap tolerance for on-device inference latency; views now stamp arrival time instead of post-inference time.
- Added
EnrollmentResult.registrationImageBytes(front-pose JPEG copy for host review-photo upload).
0.1.0 #
- Initial beta release.
- Added on-device face detection and package-owned result types.
- Replaced the unidentified MobileFaceNet pipeline with bundled, hash-pinned OpenCV YuNet detection and SFace INT8 128-dimensional recognition.
- Added only the required Apache-licensed MediaPipe landmark and blendshape models for live pose and blink guidance.
- Added versioned, JSON-serializable three-pose templates.
- Added 1:1 verification with typed compatibility and quality failures.
- Added randomized blink/head-turn basic liveness state machine.
- Added ready-made enrollment and verification camera widgets.
- Added Android/iOS example application and migration/security documentation.
- Added schema-v2 templates with secure random template IDs, strict field and compatibility parsing, three 128-value samples, and a verified centroid.
- Added bounded canonical preprocessing, deterministic 1600-pixel resize, RGB-to-BGR conversion, SFace align-crop, and four-asset integrity checks.
- Added runtime configuration and strict template validation.
- Added enrollment same-person consistency checks and per-call thresholds.
- Hardened liveness transitions, camera lifecycle handling, rotation, capture recovery, and immediate post-challenge capture.
- Removed the separate runtime permission dependency; camera initialization now owns permission requests.
- Fixed Android CameraX live detection by using its supported multi-plane YUV420 stream instead of an undecodable one-plane NV21 stream.
- Corrected selfie-facing left/right pose and liveness directions.
- Added responsive enrollment/verification widgets, expanded text theming, host error callbacks, and clearer capture guidance.
- Added stale-frame and stale-operation guards for lifecycle, template, kit, and configuration changes.
- Bounded serialized template dimensions and sanitized public processing failures so native exception details are not exposed.
- Removed
face_detection_tflite, eleven unused transitive model assets, and all legacy MobileFaceNet identifiers and pipeline code. - Added complete SFace, YuNet, and MediaPipe provenance/licence notices.
Metadata
0
likes
160
points
89
downloads
Documentation
Publisher
Weekly Downloads
Metadata
On-device face detection, guided enrollment, liveness challenges, and 1:1 verification for Flutter.
Repository (GitHub)
View/report issues
Contributing
Topics
#face-detection #face-verification #biometrics #on-device-ml #liveness
License
Apache-2.0 (license)
Dependencies
camera, crypto, flutter, flutter_litert, image, opencv_dart